Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

Comprehensive Attribute Encoding and Dynamic LSTM HyperModels for Outcome Oriented Predictive Business Process Monitoring

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper reports 100% accuracy on all outcome classes of three real loan-process logs using self-tuning LSTM hypermodels with purpose-built attribute encoding.

desk verdict Useful encoding ideas wrapped around an evaluation that does not support the 100% accuracy claim. read the letter →

arxiv 2506.03696 v2 pith:E5CA7JU2 submitted 2025-06-04 cs.LG

Fang Wang , Paolo Ceravolo , Ernesto Damiani This is my paper · ORCID
classification cs.LG
keywords predictivebusinessprocessmonitoringoutcomepredictionLSTMeventencodingpseudo-embeddingsimultaneouseventsclassimbalancehyperparametertuning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper claims that outcome prediction for ongoing business processes can be pushed to perfect accuracy by engineering the input representation rather than by making the network deeper. To do this it proposes a two-level encoding scheme: event labels are split into verb and descriptor tokens; sequence-level and event-level attributes are embedded through tf-idf pseudo-embeddings that capture attribute correlations and dynamically binned durations; simultaneous events are concatenated into multidimensional embeddings and augmented with time-difference flags. These representations feed a family of self-tuning LSTM hypermodels whose hyperparameters are selected by Hyperband. On three balanced real-world loan-process logs the paper reports 100% accuracy on every outcome class, and on an imbalanced synthetic healthcare log it reports weighted F1 above 0.86. A sympathetic reader would take the central message to be that representation engineering plus modest, automatically tuned recurrent networks is sufficient for outcome-oriented predictive process monitoring.

What carries the argument

The carrying mechanism is the encoding stack, not a theorem. Four components do the work. First, event-label featurization turns each activity label into a verb plus up to two descriptor tokens, which the paper argues reduces vocabulary sparsity. Second, a tf-idf correlation pseudo-embedding treats each sequence as a document over universal-attribute combinations and assigns each event a vector $v^{\mathrm{cor}}_i$. Third, a duration pseudo-embedding bins each event's duration $T^d_i$ by a cut-off plus quantile binning and encodes the bin with tf-idf, giving $v^{\mathrm{bin}}_i$. Fourth, simultaneous-event vectorization concatenates embeddings of co-occurring events and appends the time-difference flag $\Delta T_i = T^s_i - T^s_{i-1}$. The B/D/DC/T-LSTM architectures and their F/M/U variants fuse these vectors, concatenate the final LSTM output with the sequence-level vector, and pass the result through dense layers; Hyperband tunes depth, width, regularization, optimizer, and learning-rate schedule.

What would settle it

Take the best M-B-LSTM and F-B-LSTM configurations, truncate every BPIC12 test trace immediately before its final ACCEPTED/DECLINED/CANCELLED event, retrain on the truncated prefixes, and report per-class accuracy. If accuracy drops substantially below 1.0, the perfect scores were produced by the terminal label token being part of the input; if it stays at 1.0, the encoding alone is doing the separation.

Watch

Extended reading notes

Core claim

The paper's central claim is that sequence-level outcome prediction in predictive business process monitoring reduces to choosing the right encoding of each trace, and that the proposed encoding suite makes outcomes separable enough for a single multiclass LSTM to classify them perfectly. The authors define event labels semantically (verb plus descriptors), construct tf-idf pseudo-embedding vectors for universal-attribute combinations and for dynamically binned durations, and vectorize simultaneous events by concatenating per-event embeddings and adding $\Delta T_i = T^s_i - T^s_{i-1}$. These inputs are fed through one of several LSTM architectures that merge event-level and sequence-level representations into a final concatenated vector for softmax classification. The reported scores are all classes at accuracy 1.0 on BPIC12, BPIC12-A, and BPIC12-O, and weighted F1 above 0.86 on the imbalanced Patients dataset. The paper further claims that a single multiclass classifier matches or beats earlier systems that used three separate binary classifiers.

Load-bearing premise

The perfect-accuracy claim presupposes that the LSTM input is a genuine prefix of the trace, with the final outcome event (ACCEPTED, DECLINED, or CANCELLED) excluded; the paper never states that it truncates the input, and those outcome names are also event labels in Table III.

Editorial extensions

If this is right

  • A single multiclass LSTM hypermodel can replace the three separate binary classifiers used by earlier BPIC12 outcome-prediction systems and still report accuracy 1.0.
  • On the paper's evidence, richer encodings let simpler recurrent stacks win: the Hyperband-tuned M-B-LSTM variants use fewer LSTM layers, and F-D-LSTM uses one LSTM layer per input stream.
  • The framework transfers across balanced and imbalanced logs, reporting perfect accuracy on the BPIC12 variants and weighted F1 above 0.86 on Patients without dataset-specific architecture changes.
  • The duration-bin and time-difference components are the features credited for the gains, so logs with regular temporal structure should benefit most from these encodings rather than from deeper LSTM stacks.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • An editorial extension: rerunning the same models on prefixes truncated just before the final ACCEPTED/DECLINED/CANCELLED event would settle whether the reported perfect scores come from the encoding or from the model reading the terminal label token, a test the paper does not describe.
  • The tf-idf pseudo-embedding over attribute combinations treats each trace as a document, so a natural transfer is to other temporal outcome tasks such as churn prediction, credit-risk scoring, or hospital length-of-stay forecasting.
  • A further editorial inference is that the contribution of the time-difference flag could be isolated by an ablation that removes only that feature while keeping the rest of the pipeline fixed, which the paper itself lists as open future work.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper proposes a suite of LSTM-based hypermodels for outcome-oriented predictive business process monitoring, combining hierarchical attribute encoding, character-level featurization of event labels, pseudo-embeddings for attribute correlations and durations, and specialized handling of simultaneous events. The authors report perfect (100%) accuracy on the BPIC12, BPIC12-A, and BPIC12-O datasets and weighted F1 scores above 0.86 on the Patients dataset, and they attribute these results to their proposed encoding and embedding strategies. The core evaluation, however, is not trustworthy: the BPIC12 outcome classes are event labels that appear to be included in the LSTM input without any described truncation, and the reported performance on Patients comes from the same validation split used for hyperparameter selection. As a result, the headline empirical claims are not supported as predictions of future outcomes.

Significance. If the reported results were valid, the proposed modular encoding toolkit and LSTM hypermodel variants would constitute a useful engineering contribution to predictive process monitoring, and the authors are to be credited for releasing a code repository, using public benchmarks, and reporting computational costs. However, the central empirical claims are undermined by target leakage in the BPIC12 experiments and by an evaluation protocol that selects and evaluates models on the same validation split. The paper therefore does not currently provide evidence that the proposed encoding strategies improve outcome prediction over baselines; the significance of the work cannot be assessed from the numbers as reported.

major comments (3)
  1. [Sections V-A, VI-B.2; Table III] The BPIC12 evaluation appears to have target leakage: the outcome classes are the final event labels (accepted/approved, declined, canceled), and Table III featurizes ACCEPTED, APPROVED, and CANCELLED as ordinary event-label tokens. Section V-A states that each BPIC12 sequence concludes with one of these outcomes, and no step is described that truncates the input sequence before the final outcome event or removes that event from the LSTM input. A model consuming the full trace can therefore read the outcome from its last input token, making the reported perfect accuracy a tautology rather than a predictive result. The caveat in Section VI-B.2 about the simplicity of BPIC12-A/O does not address this leakage. The authors must specify the exact prefix truncation rule; if the final event is retained in the input, the 100% claims should be withdrawn.
  2. [Section V-C] The evaluation protocol selects and evaluates on the same validation split. The text states that Hyperband maximizes validation accuracy or weighted F1 and that final results are obtained by 'retrieving the best model directly or rebuilding it using the selected hyperparameters.' No separate held-out test set or nested cross-validation is described. Reporting the best validation performance as the final result overstates generalization and can be severely biased, especially with the large hyperparameter search space in Table I. An independent test set or nested resampling procedure is required before any of the reported performance numbers can be interpreted.
  3. [Table VII and Section VI-B.2] The comparison against prior work in Table VII is not controlled: the baseline accuracies are taken from other studies with different feature encodings, training protocols, and possibly different prediction tasks (the paper itself notes that earlier work decomposed the task into binary classifiers). No confidence intervals, standard deviations, or significance tests are provided, and the comparison is further invalidated by the target leakage described above. The authors should compare all methods under an identical prefix-based protocol with uncertainty estimates.
minor comments (4)
  1. [Table III] The header contains a typo: 'Patience Dataset' should read 'Patients Dataset.'
  2. [Table IV] The first BPIC12 entry appears corrupted: '12 5h51m27s' should presumably read 'BPIC12 5h51m27s.'
  3. [Section V-B] The 'Patients' dataset is described as synthetic, but no source or generation procedure is cited; this limits reproducibility of the benchmark.
  4. [Section V-A and Table VII] Dataset naming is inconsistent: the text and table headers use 'BPI12A/O' and 'BPIC12-A/O' interchangeably; please standardize.

Circularity Check

2 steps flagged · score 9.0 of 10

Perfect BPIC12 accuracy is target leakage: the outcome classes are featurized event labels and no trace truncation is described, so the LSTM can read the answer from the final input token.

  1. self definitional [Sections III-A, IV-A.1, V-A, Table III; result in Section VI-B.2]
    "Each sequence concludes with one of three outcomes: accepted (approved) , declined, or canceled. ... ACCEPTED accept <NO_DESC> ... APPROVED approve <NO_DESC> ... CANCELLED cancel <NO_DESC> ... The base LSTM model processes each event Xi within a sequence Sj by combining categorical (CXi, CSj) and numerical attributes (NXi, NSj) into feature vectors"

    The outcome classes are the final activity labels of the same event sequence that is fed to the LSTM. Section V-A states each BPIC12 sequence concludes with one of three outcomes; Table III featurizes ACCEPTED and CANCELLED (and analogous final activities) as ordinary event-label tokens; Section III-A defines the input as the entire sequence with no truncation; Section IV-A.1 says the LSTM processes every event Xi in Sj. Thus the final input token can carry the true class itself, and a model that copies the last activity label into its softmax output attains accuracy 1.0 in Table VII by construction. The paper describes no prefix-truncation rule, so the reported perfect accuracy is target leakage rather than an independent predictive result.

  2. fitted input called prediction [Section V-C (LSTM Hyperparameters Searching)]
    "Each model was trained using an 80/20 train-validation split, and early stopping was employed to prevent overfitting. After the tuning process, optimal hyperparameters were extracted from the best-performing trial. Final evaluation results were obtained by either retrieving the best model directly or rebuilding it using the selected hyperparameters."

    Only an 80/20 train-validation split is described; no held-out test set is mentioned. Hyperparameters are selected by maximizing validation accuracy or weighted F1, and the final evaluation results are obtained by retrieving the best trial or rebuilding with those selected hyperparameters. With no other split defined, the reported numbers are the very criterion used for model selection, so they are selected rather than independently predicted. This makes the headline comparison statistically forced by construction and removes the independence needed to support the claimed 100% accuracy.

full rationale

The central empirical claim of the paper is that M-B-LSTM, F-B-LSTM, and F-D-LSTM reach accuracy 1.0 on BPIC12 and BPIC12-A/O (Section VI-B.2, Table VII). That claim reduces to the input by definition: BPIC12 sequences are said to conclude with one of three outcomes, and the same outcome-defining activity labels (ACCEPTED, APPROVED, CANCELLED, and analogous final activities such as DECLINED) are included in Table III as ordinary event tokens processed by the LSTM. The paper never states that the final outcome event is excluded from the prefix, so the model can simply read the last input token to obtain the class label. This is not a fitted-parameter issue but a definitional circularity: the outcome the model is asked to predict is part of the event sequence it is given. The comparison against prior 0.63-0.69 baselines is therefore invalid under this protocol, since those benchmarks predict before the final outcome event occurs. A secondary but reinforcing issue is that the evaluation protocol uses the validation split for hyperparameter selection and then reports the best trial as the final result, with no independent test split described. The paper's own caveat that BPIC12-A/O simplicity may inflate performance does not address this leakage. Because the headline perfect accuracy is the load-bearing result and it is achieved by construction, the circularity score is high; the encoding and architecture contributions may still be non-circular in themselves, but the validation of their central claim is not.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

No new physical entities are postulated. The free parameters are design choices in the encoding and hyperparameter search; the most consequential assumption is that the outcome event can be included in the input without invalidating the prediction task.

free parameters (4)
  • Duration binning cutoff Tcut = 5 minutes for Patients, 0 for BPIC12 variants
    Chosen by hand to balance bin frequencies (Algorithm 2, Section V-B). Affects the duration pseudo-embedding that D-LSTM and DC-LSTM rely on.
  • Number of duration quantile bins Nb* = Implied 24 total bins for Patients; 2 bins for BPIC12
    Manual selection in Section V-B; controls granularity of duration representation.
  • LSTM/dense hyperparameters (units, layers, dropout, L2, LR schedule) = Per Table V; selected by Hyperband on validation
    The 'HyperModel' claim rests on this search; no separate test set is used to validate the selection.
  • Dummy attribute for correlation pseudo-embedding = One dummy categorical attribute added to Patients
    Section V-B states a dummy attribute was added to enable DC-LSTM processing; this is an ad hoc input modification.
assumptions (4)
  • domain assumption The full trace including the final outcome event is a valid input for outcome prediction
    Sections III-A and V-A define the input as the entire sequence; for BPIC12, the final event label is the outcome (Table III). The paper does not describe any prefix truncation, so it implicitly assumes the model may see the answer.
  • domain assumption Manual verb-descriptor featurization covers all activity labels correctly
    Table III human annotations map labels to verb and descriptor tokens; if a label is mis-decomposed, the embedding is corrupted.
  • ad hoc to paper Tf-idf scores computed on event sequences capture predictive attribute correlations
    The pseudo-embedding methods (Algorithms 1 and 2) assume document-level tf-idf weighting is informative for sequential outcome prediction.
  • ad hoc to paper Validation accuracy/F1 after Hyperband search generalizes to unseen traces
    Section V-C reports final results from the best trial on the 80/20 validation split, with no separate test set.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Comprehensive Attribute Encoding and Dynamic LSTM HyperModels for Outcome Oriented Predictive Business Process Monitoring." pith.science (2026). https://pith.science/paper/E5CA7JU2

@misc{pith2026250603696,
  author       = {Pith},
  title        = {Pith review of: Comprehensive Attribute Encoding and Dynamic LSTM HyperModels for Outcome Oriented Predictive Business Process Monitoring},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/E5CA7JU2}},
  note         = {Machine review of arXiv:2506.03696}
}
read the original abstract

Predictive Business Process Monitoring (PBPM) aims to forecast future outcomes of ongoing business processes. However, existing methods often lack flexibility to handle real-world challenges such as simultaneous events, class imbalance, and multi-level attributes. While prior work has explored static encoding schemes and fixed LSTM architectures, they struggle to support adaptive representations and generalize across heterogeneous datasets. To address these limitations, we propose a suite of dynamic LSTM HyperModels that integrate two-level hierarchical encoding for event and sequence attributes, character-based decomposition of event labels, and novel pseudo-embedding techniques for durations and attribute correlations. We further introduce specialized LSTM variants for simultaneous event modeling, leveraging multidimensional embeddings and time-difference flag augmentation. Experimental validation on four public and real-world datasets demonstrates up to 100% accuracy on balanced datasets and F1 scores exceeding 86\% on imbalanced ones. Our approach advances PBPM by offering modular and interpretable models better suited for deployment in complex settings. Beyond PBPM, it contributes to the broader AI community by improving temporal outcome prediction, supporting data heterogeneity, and promoting explainable process intelligence frameworks.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. HGCN(O): A Self-Tuning GCN HyperModel Toolkit for Outcome Prediction in Event-Sequence Data

    cs.LG 2025-07 reject novelty 3.0 of 10

    A self-tuning toolkit of four GCN input architectures (O-, T-, TP-, TE-GCN) is applied to outcome prediction; the claimed advantage over baselines is unsupported, and the balanced-data perfect scores are likely leakag...

Reference graph

Works this paper leans on

62 extracted references · 53 canonical work pages · cited by 1 Pith paper

  1. [1]

    Tuning machine learning to address process mining requirements,

    P. Ceravolo, S. B. Junior, E. Damiani, and W. Van Der Aalst, “Tuning machine learning to address process mining requirements,” IEEE Access, vol. 12, pp. 24 583–24 595, 2024

    work page 2024

  2. [2]

    A general process mining framework for correlating, predicting and clustering dynamic behavior based on event logs,

    M. De Leoni, W. M. Van Der Aalst, and M. Dees, “A general process mining framework for correlating, predicting and clustering dynamic behavior based on event logs,” Information Systems , vol. 56, pp. 235– 257, 2016

    work page 2016

  3. [3]

    Clustering-based predictive process monitoring,

    C. Di Francescomarino, M. Dumas, F. M. Maggi, and I. Teinemaa, “Clustering-based predictive process monitoring,” IEEE transactions on services computing, vol. 12, no. 6, pp. 896–909, 2016

    work page 2016

  4. [4]

    Complex symbolic sequence encodings for predictive monitoring of business processes,

    A. Leontjeva, R. Conforti, C. Di Francescomarino, M. Dumas, and F. M. Maggi, “Complex symbolic sequence encodings for predictive monitoring of business processes,” in Business Process Management: 13th International Conference, BPM 2015, Innsbruck, Austria, August 31–September 3, 2015, Proceedings 13 . Springer, 2015, pp. 297–313

    work page 2015

  5. [5]

    Predictive analytics for semi-structured case oriented business processes,

    G. T. Lakshmanan, S. Duan, P. T. Keyser, F. Curbera, and R. Kha- laf, “Predictive analytics for semi-structured case oriented business processes,” in Business Process Management Workshops: BPM 2010 International Workshops and Education Track, Hoboken, NJ, USA, September 13-15, 2010, Revised Selected Papers 8 . Springer, 2011, pp. 640–651

    work page 2010

  6. [6]

    Enhancing predictive process monitoring with time-related feature engineering,

    R. S. Oyamada, G. M. Tavares, S. B. Junior, and P. Ceravolo, “Enhancing predictive process monitoring with time-related feature engineering,” in Advanced Information Systems Engineering , G. Guizzardi, F. Santoro, H. Mouratidis, and P. Soffer, Eds

    work page

  7. [7]

    Trace encoding in process mining: A survey and benchmarking,

    G. M. Tavares, R. S. Oyamada, S. B. Junior, and P. Ceravolo, “Trace encoding in process mining: A survey and benchmarking,” Engineering Applications of Artificial Intelligence , vol. 126, p. 107028, 2023

    work page 2023

  8. [8]

    Predicting process behaviour using deep learning,

    J. Evermann, J.-R. Rehse, and P. Fettke, “Predicting process behaviour using deep learning,” Decision Support Systems , vol. 100, pp. 129–140, 2017

    work page 2017

Show all 62 references
  1. [9]

    Predictive business process monitoring with lstm neural networks,

    N. Tax, I. Verenich, M. La Rosa, and M. Dumas, “Predictive business process monitoring with lstm neural networks,” in Advanced Informa- tion Systems Engineering: 29th International Conference, CAiSE 2017, Essen, Germany, June 12-16, 2017, Proceedings 29 . Springer, 2017, pp. 477–492

    2017

  2. [10]

    Predictive monitoring of business processes,

    F. M. Maggi, C. Di Francescomarino, M. Dumas, and C. Ghidini, “Predictive monitoring of business processes,” in Advanced Informa- tion Systems Engineering: 26th International Conference, CAiSE 2014, Thessaloniki, Greece, June 16-20, 2014. Proceedings 26 . Springer, 2014, pp. 457–472

    2014

  3. [11]

    Evaluating and predicting overall process risk using event logs,

    A. Pika, W. M. van der Aalst, M. T. Wynn, C. J. Fidge, and A. H. ter Hofstede, “Evaluating and predicting overall process risk using event logs,” Information Sciences, vol. 352, pp. 98–120, 2016

    2016

  4. [12]

    Outcome- oriented predictive process monitoring: Review and benchmark,

    I. Teinemaa, M. Dumas, M. L. Rosa, and F. M. Maggi, “Outcome- oriented predictive process monitoring: Review and benchmark,” ACM Transactions on Knowledge Discovery from Data (TKDD), vol. 13, no. 2, pp. 1–57, 2019

    2019

  5. [13]

    Pre- dicting critical behaviors in business process executions: when evidence counts,

    L. Genga, C. Di Francescomarino, C. Ghidini, and N. Zannone, “Pre- dicting critical behaviors in business process executions: when evidence counts,” in Business Process Management Forum: BPM Forum 2019, Vienna, Austria, September 1–6, 2019, Proceedings 17. Springer, 2019, pp. ...

    2019

  6. [14]

    Predictive process monitoring,

    C. Di Francescomarino and C. Ghidini, “Predictive process monitoring,” in Process mining handbook. Springer International Publishing Cham, 2022, pp. 320–346

    2022

  7. [15]

    Orange: outcome-oriented predictive process monitoring based on image encoding and cnns,

    V . Pasquadibisceglie, A. Appice, G. Castellano, D. Malerba, and G. Modugno, “Orange: outcome-oriented predictive process monitoring based on image encoding and cnns,” IEEE Access, vol. 8, pp. 184 073– 184 086, 2020

    2020

  8. [16]

    Specification-driven multi-perspective predictive business process monitoring,

    A. Santoso, “Specification-driven multi-perspective predictive business process monitoring,” in Enterprise, Business-Process and Information Systems Modeling: 19th International Conference, BPMDS 2018, 23rd International Conference, EMMSAD 2018, Held at CAiSE 2018, Tallinn, Es...

    2018

  9. [17]

    Improving business process quality through exception understanding, prediction, and pre- vention,

    D. Grigori, F. Casati, U. Dayal, and M.-C. Shan, “Improving business process quality through exception understanding, prediction, and pre- vention,” in Proceedings of the 27th International Conference on Very Large Data Bases , 2001, pp. 159–168

    2001

  10. [18]

    Business process intelligence,

    D. Grigori, F. Casati, M. Castellanos, U. Dayal, M. Sayal, and M.-C. Shan, “Business process intelligence,” Computers in industry , vol. 53, no. 3, pp. 321–343, 2004

    2004

  11. [19]

    Predictive business operations management,

    M. Castellanos, N. Salazar, F. Casati, U. Dayal, and M.-C. Shan, “Predictive business operations management,” in Proceedings of the 4th international conference on Databases in Networked Information Systems, 2005, pp. 1–14

    2005

  12. [20]

    Intra and inter-case features in predictive process mon- itoring: A tale of two dimensions,

    A. Senderovich, C. Di Francescomarino, C. Ghidini, K. Jorbina, and F. M. Maggi, “Intra and inter-case features in predictive process mon- itoring: A tale of two dimensions,” in Business Process Management: 15th International Conference, BPM 2017, Barcelona, Spain, September 10...

    2017

  13. [21]

    Periodic performance prediction for real-time business process monitoring,

    B. Kang, D. Kim, and S.-H. Kang, “Periodic performance prediction for real-time business process monitoring,” Industrial Management & Data Systems, vol. 112, no. 1, pp. 4–23, 2012

    2012

  14. [22]

    Deep learning for predictive business process monitoring: Review and benchmark,

    E. Rama-Maneiro, J. C. Vidal, and M. Lama, “Deep learning for predictive business process monitoring: Review and benchmark,” IEEE Transactions on Services Computing, vol. 16, no. 1, pp. 739–756, 2021

    2021

  15. [23]

    Deep learn- ing process prediction with discrete and continuous data features,

    S. Sch ¨onig, R. Jasinski, L. Ackermann, and S. Jablonski, “Deep learn- ing process prediction with discrete and continuous data features,” in Proceedings of the 13th international conference on evaluation of novel approaches to software engineering , 2018, pp. 314–319

    2018

  16. [24]

    Lstm networks for data-aware remaining time prediction of business process instances,

    N. Navarin, B. Vincenzi, M. Polato, and A. Sperduti, “Lstm networks for data-aware remaining time prediction of business process instances,” in 2017 IEEE Symposium Series on Computational Intelligence (SSCI) . IEEE, 2017, pp. 1–7

    2017

  17. [25]

    Learning accurate lstm models of business processes,

    M. Camargo, M. Dumas, and O. Gonz ´alez-Rojas, “Learning accurate lstm models of business processes,” in Business Process Management: 17th International Conference, BPM 2019, Vienna, Austria, September 1–6, 2019, Proceedings 17 . Springer, 2019, pp. 286–302

    2019

  18. [26]

    Ham-net: Predictive business process monitoring with a hierarchical attention mechanism,

    A. Jalayer, M. Kahani, A. Pourmasoumi, and A. Beheshti, “Ham-net: Predictive business process monitoring with a hierarchical attention mechanism,” Knowledge-Based Systems, vol. 236, p. 107722, 2022

    2022

  19. [27]

    Time matters: Time-aware lstms for predictive business process monitoring,

    A. Nguyen, S. Chatterjee, S. Weinzierl, L. Schwinn, M. Matzner, and B. Eskofier, “Time matters: Time-aware lstms for predictive business process monitoring,” in Process Mining Workshops: ICPM 2020 Inter- national Workshops, Padua, Italy, October 5–8, 2020, Revised Selected Pap...

    2020

  20. [28]

    Mm-pred: A deep predictive model for multi-attribute event sequence,

    L. Lin, L. Wen, and J. Wang, “Mm-pred: A deep predictive model for multi-attribute event sequence,” in Proceedings of the 2019 SIAM international conference on data mining . SIAM, 2019, pp. 118–126

    2019

  21. [29]

    Harane and S

    N. Harane and S. Rathi, Comprehensive Survey on Deep Learning Approaches in Predictive Business Process Monitoring . Cham: Springer International Publishing, 2020, pp. 115–128. [Online]. Available: https://doi.org/10.1007/978-3-030-38445-6 9

    2020 doi

  22. [30]

    Survey and cross-benchmark comparison of remaining time prediction methods in business process monitoring,

    I. Verenich, M. Dumas, M. L. Rosa, F. M. Maggi, and I. Teinemaa, “Survey and cross-benchmark comparison of remaining time prediction methods in business process monitoring,” ACM Transactions on Intelli- gent Systems and Technology (TIST) , vol. 10, no. 4, pp. 1–34, 2019

    2019

  23. [31]

    Text-aware predictive monitoring of business processes,

    M. Pegoraro, M. S. Uysal, D. B. Georgi, and W. M. van der Aalst, “Text-aware predictive monitoring of business processes,” in Business Information Systems, 2021, pp. 221–232

    2021

  24. [32]

    A systematic literature review on state-of-the-art deep learning methods for process prediction,

    D. A. Neu, J. Lahann, and P. Fettke, “A systematic literature review on state-of-the-art deep learning methods for process prediction,” Artificial Intelligence Review, vol. 55, no. 2, pp. 801–827, 2022

    2022

  25. [33]

    Predictive business process monitoring with structured and unstructured data,

    I. Teinemaa, M. Dumas, F. M. Maggi, and C. Di Francescomarino, “Predictive business process monitoring with structured and unstructured data,” in Business Process Management: 14th International Conference, BPM 2016, Rio de Janeiro, Brazil, September 18-22, 2016. Proceedings

    2016

  26. [34]

    Springer, 2016, pp. 401–417

    2016

  27. [35]

    Comparing and combining pre- dictive business process monitoring techniques,

    A. Metzger, P. Leitner, D. Ivanovi ´c, E. Schmieders, R. Franklin, M. Carro, S. Dustdar, and K. Pohl, “Comparing and combining pre- dictive business process monitoring techniques,” IEEE Transactions on Systems, Man, and Cybernetics: Systems , vol. 45, no. 2, pp. 276–290, 2014

    2014

  28. [36]

    Genetic algorithms for hyperparam- eter optimization in predictive business process monitoring,

    C. Di Francescomarino, M. Dumas, M. Federici, C. Ghidini, F. M. Maggi, W. Rizzi, and L. Simonetto, “Genetic algorithms for hyperparam- eter optimization in predictive business process monitoring,”Information Systems, vol. 74, pp. 67–83, 2018

    2018

  29. [37]

    Outcome-oriented predictive pro- cess monitoring with attention-based bidirectional lstm neural networks,

    J. Wang, D. Yu, C. Liu, and X. Sun, “Outcome-oriented predictive pro- cess monitoring with attention-based bidirectional lstm neural networks,” in 2019 IEEE International Conference on Web Services (ICWS). IEEE, 2019, pp. 360–367

    2019

  30. [38]

    Learning effective neural nets for outcome prediction from partially labelled log data,

    F. Folino, G. Folino, M. Guarascio, and L. Pontieri, “Learning effective neural nets for outcome prediction from partially labelled log data,” in 2019 IEEE 31st International Conference on Tools with Artificial Intelligence (ICTAI). IEEE, 2019, pp. 1396–1400

    2019

  31. [39]

    Classifying process instances using recurrent neural networks,

    M. Hinkka, T. Lehto, K. Heljanko, and A. Jung, “Classifying process instances using recurrent neural networks,” in Business Process Man- agement Workshops: BPM 2018 International Workshops, Sydney, NSW, Australia, September 9-14, 2018, Revised Papers 16 . Springer, 2019, pp. 313–324

    2018

  32. [40]

    Pustejovsky and A

    J. Pustejovsky and A. Stubbs, Natural Language Annotation for Machine Learning: A guide to corpus-building for applications . ” O’Reilly Media, Inc.”, 2012

    2012

  33. [41]

    Change patterns and change support features–enhancing flexibility in process-aware infor- mation systems,

    B. Weber, M. Reichert, and S. Rinderle-Ma, “Change patterns and change support features–enhancing flexibility in process-aware infor- mation systems,” Data & knowledge engineering , vol. 66, no. 3, pp. 438–466, 2008

    2008

  34. [42]

    Graves, Supervised sequence labelling

    A. Graves, Supervised sequence labelling . Springer, 2012

    2012

  35. [43]

    An exploration of dropout with lstms

    G. Cheng, V . Peddinti, D. Povey, V . Manohar, S. Khudanpur, and Y . Yan, “An exploration of dropout with lstms.” 2017

    2017

  36. [44]

    A theoretically grounded application of dropout in recurrent neural networks,

    Y . Gal and Z. Ghahramani, “A theoretically grounded application of dropout in recurrent neural networks,” Advances in neural information processing systems, vol. 29, 2016

    2016

  37. [45]

    Neural networks for machine learning, lecture 6a overview of mini-batch gradient descent,

    G. Hinton, N. Srivastava, and K. Swersky, “Neural networks for machine learning, lecture 6a overview of mini-batch gradient descent,” 2012

    2012

  38. [46]

    Batch normalization: Accelerating deep network training by reducing internal covariate shift,

    S. Ioffe and C. Szegedy, “Batch normalization: Accelerating deep network training by reducing internal covariate shift,” in International conference on machine learning . pmlr, 2015, pp. 448–456

    2015

  39. [47]

    Deep sparse rectifier neural networks,

    X. Glorot, A. Bordes, and Y . Bengio, “Deep sparse rectifier neural networks,” in Proceedings of the fourteenth international conference on artificial intelligence and statistics . JMLR Workshop and Conference Proceedings, 2011, pp. 315–323

    2011

  40. [48]

    Methods for interpreting and understanding deep neural networks,

    G. Montavon, W. Samek, and K.-R. M ¨uller, “Methods for interpreting and understanding deep neural networks,” Digital signal processing , vol. 73, pp. 1–15, 2018

    2018

  41. [49]

    Deep learning of representations: Looking forward,

    Y . Bengio, “Deep learning of representations: Looking forward,” in International conference on statistical language and speech processing . Springer, 2013, pp. 1–37

    2013

  42. [50]

    On the expressive power of deep neural networks,

    M. Raghu, B. Poole, J. Kleinberg, S. Ganguli, and J. Sohl-Dickstein, “On the expressive power of deep neural networks,” in international conference on machine learning . PMLR, 2017, pp. 2847–2854

    2017

  43. [51]

    On the importance of initialization and momentum in deep learning,

    I. Sutskever, J. Martens, G. Dahl, and G. Hinton, “On the importance of initialization and momentum in deep learning,” in International conference on machine learning . PMLR, 2013, pp. 1139–1147

    2013

  44. [52]

    Llr: Learning learning rates by lstm for training neural networks,

    C. Yu, X. Qi, H. Ma, X. He, C. Wang, and Y . Zhao, “Llr: Learning learning rates by lstm for training neural networks,” Neurocomputing, vol. 394, pp. 41–50, 2020

    2020

  45. [53]

    A comparison of lstm and gru networks for learning symbolic sequences,

    R. Cahuantzi, X. Chen, and S. G ¨uttel, “A comparison of lstm and gru networks for learning symbolic sequences,” in Science and Information Conference. Springer, 2023, pp. 771–785

    2023

  46. [54]

    Bert: Pre-training of deep bidirectional transformers for language understanding,

    J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “Bert: Pre-training of deep bidirectional transformers for language understanding,” in Pro- ceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technolo...

    2019

  47. [55]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980 , 2014

    2014 arXiv

  48. [56]

    Two-layer intelligent learn- ing control using output recurrent fuzzy neural lstm-bls with rmsprop,

    A. Rospawan, C.-C. Tsai, and C.-C. Hung, “Two-layer intelligent learn- ing control using output recurrent fuzzy neural lstm-bls with rmsprop,” IEEE Access, 2025

    2025

  49. [57]

    Glove: Global vectors for word representation,

    J. Pennington, R. Socher, and C. D. Manning, “Glove: Global vectors for word representation,” in Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP) , 2014, pp. 1532–1543. 12

    2014

  50. [58]

    Efficient estimation of word representations in vector space,

    T. Mikolov, K. Chen, G. Corrado, and J. Dean, “Efficient estimation of word representations in vector space,” arXiv preprint arXiv:1301.3781 , 2013

    2013 arXiv

  51. [59]

    Lstm hyper-parameter selection for malware detection: Interaction effects and hierarchical selection approach,

    M. Sewak, S. K. Sahay, and H. Rathore, “Lstm hyper-parameter selection for malware detection: Interaction effects and hierarchical selection approach,” in 2021 International Joint Conference on Neural Networks (IJCNN). IEEE, 2021, pp. 1–9

    2021

  52. [60]

    Bpi challenge 2012,

    B. Van Dongen, “Bpi challenge 2012,” 2012. [Online]. Available: https://data.4tu.nl/articles/ /12689204/1

    2012

  53. [61]

    Outcome-oriented prescriptive process monitoring based on temporal logic patterns,

    I. Donadello, C. Di Francescomarino, F. M. Maggi, F. Ricci, and A. Shikhizada, “Outcome-oriented prescriptive process monitoring based on temporal logic patterns,” Engineering Applications of Artificial Intel- ligence, vol. 126, p. 106899, 2023

    2023

  54. [62]

    Process outcome prediction: Cnn vs. lstm (with attention),

    H. Weytjens and J. De Weerdt, “Process outcome prediction: Cnn vs. lstm (with attention),” in Business Process Management Workshops: BPM 2020 International Workshops, Seville, Spain, September 13–18, 2020, Revised Selected Papers 18 . Springer, 2020, pp. 321–333

    2020

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.